AIAGENTDEV
AIAGENTDEV · AIAGENTDEV Workspace
What can we train today?
Crucible Continuum
The AIAGENTDEV Workspace · A model foundry

The enterprise model foundry.

AIAGENTDEV Workspace is the environment where an organisation builds its own models on its own resource. Crucible forges those models into Ingots. Continuum takes the Ingots and makes them shared across the whole organisation. The outcome is a private model, hosted on AIAGENTDEV.

01 · Forge

Crucible makes Ingots

Crucible forges a private model from the organisation's own resource, cast as an Ingot: a versioned, owned model artifact.

02 · Share

Continuum shares them

Continuum takes each Ingot and makes it shared across the whole organisation, one model every team and agent can draw on.

03 · Host

Hosted on AIAGENTDEV

The outcome is a private model, hosted and served live on AIAGENTDEV compute, secured inside the organisation's own boundary.

How you work with it

As easy as chatting with Lyra.

Working with organisational data no longer needs a data team and a ticket queue. Ask AIAGENTDEV Workspace in chat, speak to it out loud, hand a job to an agent, or call the API. However your people work, every route reaches the same workspace, on your own data and inside your own boundary.

Chattyping
Summarise this week’s support tickets.
1,204 tickets across 3 themes. Billing 41%, onboarding 22%, latency 14%. Draft reply ready.
Message AIAGENTDEV Workspace…Summarise this week’s support tickets.
Speechlistening
“Cut our resolution time in half.”
Agentsrunning
Pull this week’s ticketsdone
Cluster the recurring themesdone
Draft the response playbookdone
Notify the support leaddone
API200 OK
POST /v1/foundry/run   { "source": "support_tickets",     "action": "resolve" } → 200 · model.ready   { "resolved": 1204, "themes": 3 }
An example AIAGENTDEV Workspace Illustrative · Acme Corp · Support Intelligence
Acme Corp/Support Intelligence/Pipeline
Enterprise Resource
Docs · tickets · systems
Connected
Crucible
Forge · fine-tune
Forging
Ingot
acme-support
v3 · owned
Continuum
Shared across org
38 teams
Hosted on AIAGENTDEV
Private model · live
Live · 42ms
Drag to connect · click a step to inspect it
crucible.yamlcontinuum.pyserve.py
1# Crucible - forge an Ingot from enterprise resource
2ingot: acme-support
3base_model: "aiagentdev/base-mixtral-8x7b"
4resource: connect("acme/support/*")
5train:
6  method: fine_tune # base → fine-tune → distil
7  epochs: 3
8  boundary: confidential # weights never leave
9share:
10  continuum: org("acme") # shared across the org
11  scope: all_teams
12host:
13  on: aiagentdev_cloud # private model, hosted on AIAGENTDEV
14  endpoint: "/v1/chat"
15  realtime: true

What this shows. The foundry, end to end, in one AIAGENTDEV Workspace: enterprise resource is forged by Crucible into a private, owned Ingot; Continuum takes that Ingot and makes it shared across the whole organisation; and the outcome is a private model hosted on AIAGENTDEV, served live. Toggle Guided and Expert to see the same pipeline as a drag-and-drop canvas or the code beneath it. Click any step to inspect it.

Video · 16:9 Video · full-bleed 2560 × 1440 px Product film: the foundry in motion. Forge in Crucible, share through Continuum, serve live on AIAGENTDEV.
The capabilities
01 · Build · Crucible

Crucible forges a library of expert models.

Crucible is the model foundry inside AIAGENTDEV Workspace. It bolts into the organisation's own data sources, forges what it finds into small, domain-specific expert models, and hands each one back as an owned Ingot. Over time an enterprise builds a whole library of them: one per task, department or workflow, each easy to govern because it is small, specialised and does one job well.

Ingest · bolt in any source
SharePoint Google Drive Amazon S3 Postgres Salesforce Confluence Zendesk Snowflake
Connect as many sources as the task needs. Crucible ingests from all of them at once, inside the boundary.
→
Crucible
base → fine-tune → distil
→
Library of Ingots
acme-support
Support · v3
Live
acme-triage
Ops · v1
Staged
acme-billing
Finance · v2
Forging
acme-legal
Legal · v4
Live
A growing library of owned, versioned expert models.

Small, expert, governable

Each Ingot is small, domain-specific and does one enterprise task well, which makes it far easier to govern, audit and trust than one giant general model.

Continually refined

Ingots are versioned and improved over time. Feed new data or production signal back into Crucible and re-forge, so the model gets better with use.

Handed back as API or UI

Every finished Ingot is returned as a secure API endpoint to call directly, or as a UI inside the AIAGENTDEV Workspace, ready to use.

Shared with teams

Publish an Ingot to departments and teams across the organisation through Continuum, governed and permissioned, so everyone works from the same expert model.

Crucible · the forge

Watch an Ingot take shape.

Connect a source, pick a base, and forge. Crucible casts your data into a small, owned Ingot, versioned and ready to serve.

4:3 Image · half slide 1280 × 960 px Screenshot: the Crucible forge and the growing library of Ingots.
Glossary

What is an Ingot?

Ingot is our name for what Crucible produces: a small, specialised model forged from an organisation's own data. Crucible takes a large model's capability for a domain and distils it into something a fraction of the size. Because it is small it is fast and cheap to run; because it is specialised it is accurate on the organisation's own work; and because it is theirs it runs inside their own boundary and is never retired out from under them. The Crucible forges the Ingot.

Small & dense · a fraction of a frontier model's size Cheap & fast · cost per token collapses Specialised · trained on your data, accurate on your work Owned & private · yours, in your boundary, never retired
02 · Share · Continuum

Continuum shares them across the organisation.

Continuum takes an Ingot and makes it shared across the whole organisation. One private model, and the one live context it runs on, drawn on and added to by everyone across the organisation as they work.

What one team teaches the model, the whole organisation inherits at once. The Ingot stops being a single artifact and becomes shared organisational capability, owned and held inside the workspace.

The outcome

A private model, hosted on AIAGENTDEV.

Built on the organisation's own resource, shared across every team, and served live from inside its own boundary. One model the organisation owns, and no byte of it leaves their control.

Enterprise resource→ Crucible→ Ingot→ Continuum→ Private model on AIAGENTDEV
The wedge · Security & sovereignty

Built, shared and served inside your own boundary.

The whole foundry runs inside a sealed boundary. Training and serving happen on AIAGENTDEV compute with confidential, air-gapped isolation, so the organisation's data and the model's weights provably never leave its control. This is the capability neither an inference API nor a fine-tuning API can offer.

Confidential compute

Weights and data never leave

Crucible trains and Continuum serves inside an attested confidential boundary (Hyperband, NSX-in-hypervisor). No byte of resource, context or weights crosses the perimeter.

  • Air-gapped and sovereign training and serving
  • The organisation owns the Ingot and its weights outright
  • Provenance and permissions tracked across Continuum
  • Audit-grade and regulator-ready by construction
How you build in AIAGENTDEV Workspace

Two modes over one workspace

The same workspace, two ways to work. A product lead can assemble a pipeline by hand, and an engineer can open the code beneath it, on the same live workspace.

One live workspace GuidedExpert · in sync
Guided

Build by drag and drop

Source Crucible Continuum

A visual canvas where anyone on the team assembles the workspace by hand, without writing a line of code.

  • Visual pipeline for resource, forging and sharing
  • Point-and-click model forging and evaluation
  • Templates for common enterprise patterns
  • Every action maps to a real API call underneath
same data
Expert

Manage the underlying code

ingot = crucible.forge( base="llama-3.1-8b", data=source("support_tickets"), ) ingot.share(scope="team").serve()

Drop beneath the canvas to the code that runs it. Forging, sharing scope and serving are open, editable and yours to shape.

  • Full access to forging and serving configuration
  • Custom code, dependencies and orchestration
  • Reproducible, version-controlled workspaces
  • Guided and Expert edit the same workspace, live
APIEvery capability callable as a secure endpoint DashboardThe same run, watched live in the Workspace UI
How it deploys

Primary home: AIAGENTDEV Cloud. Closer to you when it's needed.

The whole workspace lives in AIAGENTDEV's own cloud environment by default. Where a specific need calls for it, we bring it closer: deployed into your own environment and onto your own hardware, or, under licence, onto third-party clouds that AIAGENTDEV deploys and maintains.

Primary

AIAGENTDEV Cloud

The default and primary home for the AIAGENTDEV Workspace. Everything, Crucible, Continuum and serving, runs fully managed on AIAGENTDEV compute.

  • Fully managed, nothing to run
  • Confidential boundary by default
  • The fastest path to a live Ingot
Direct deployment

Your own environment

For specific needs, AIAGENTDEV Workspace deploys directly into your own environment, on your own hardware, or onto a public cloud where the use case demands it.

  • Security-sensitive environments
  • Regulated markets
  • Exceptionally demanding workspaces
Licensed third-party cloud

AIAGENTDEV-operated, under licence

AIAGENTDEV's platform can also run on other third-party cloud operators under a licence agreement. In every case, AIAGENTDEV deploys and maintains it.

  • Runs on the operator of your choice
  • Deployed and maintained by AIAGENTDEV
  • Governed by a licence agreement
Considering a deployment beyond AIAGENTDEV Cloud? Speak to our partnerships team for what is possible for your use case.
Image · 21:9 Image · full-bleed 2560 × 1080 px Closing visual: the AIAGENTDEV boundary where models are built, shared and served.